面向多采样间隔大规模GPS浮动车的并行地图匹配方法研究

The existing map matching methods of floating car are primarily targeted at high-sampling-rate, by which methods it is difficult to simultaneously guarantee the accuracy and instantaneity for the map matching of large-scale floating car. Therefore, with the objective of real-time matching large-scale floating car, this project launches a study on developing a parallel map matching approach for multi-sampling-rate and large-scale GPS-based floating car by employing the matching strategy combining the global matching and the local matching as well as the distributed parallel computing technology. First of all, the rapid map matching of high-sampling-rate floating car is implemented by establishing the incremental trajectory shape similarity measure that is integrated with the hierarchy organization of the road network data. Secondly, a colony map matching method of low-sampling-rate floating car taking both of the real-time time/speed characteristics of the road sections and the floating car historical tracking information into account is exploited by means of the global optimization belief for the shortest path search. More importantly, a functional granularity partition and parallelization method for the map matching of large-scale floating car is thoroughly investigated in this project, which has been coupled with the distributed shared memory parallel computing model by using the common multi-core PC parallel cluster complemented with a reasonable partition strategy of the floating car data. Finally, a compression storage scheme concerning the floating car matching results based on the road section and linear referencing system is proposed. On this basis, it will provide a significant support for the releasing of the real-time traffic information, the traffic forecasting, the dynamic navigation, the emergency responses and some other dynamic traffic information services which are all based on the floating car technology.

本研究以大规模GPS浮动车的实时地图匹配为目标，针对当前地图匹配方法难以确保多采样间隔、大样本数据集环境下浮动车匹配的准确性和现势性问题，采用全局匹配与局部匹配并举的原则和分布式并行计算技术，开展多采样间隔大规模GPS浮动车的并行地图匹配方法研究。通过构建增量轨迹形状相似性测度，结合道路网数据的分层分级组织，实现短采样间隔浮动车的快速地图匹配；综合考虑路段实时的时间（速度）特征和浮动车历史轨迹信息，借助全局寻优思路和其他浮动车的交通行为特征，建立长采样间隔浮动车的群体地图匹配方法；在此基础上，通过对浮动车数据的合理拆分，并利用普通多核PC并行机群，研究面向大规模浮动车地图匹配的并行计算方法，以达到与分布式共享存储并行计算模型的高度耦合；提出面向路段基于线性参照坐标的匹配结果压缩存储方案。为后继基于浮动车技术的实时路况发布、交通预测、动态导航及紧急事故响应等实时动态交通信息服务奠定重要基础。

本项目针对多采样间隔大规模GPS浮动车轨迹数据的快速地图匹配展开了一系列研究工作。采用多级道路网模型组织道路网数据，从而加快浮动车匹配过程中相邻采样间隔内对拓扑邻接路段的读取速度。针对浮动车实时地图匹配，提出了一种面向大规模浮动车数据处理的离线计算与在线匹配相结合的实时地图匹配方法，将实时性能优秀的局部匹配法分解为在线匹配与离线计算两部分，借助多级道路网网格索引，利用预计算技术，预存储匹配结果，大幅提高在线匹配效率。针对长采样间隔难题，综合考虑浮动车轨迹点之间的整体特性，在局部和全局地图匹配算法基础上，给出了一种基于改进 AOE 网络的浮动车数据地图匹配方法。针对多采样间隔难题，通过对浮动车历史轨迹的分析和挖掘，给出了一种路段历史轨迹上下文的概念，根据各个路段的历史轨迹上下文实现不同采样间隔下的地图匹配。为进一步提升浮动车地图匹配效率，选择MPI作为并行计算开发环境，采用主从模式对多采样间隔下的浮动车地图匹配算法进行了并行化改造。实验结果表明，在2台普通多核PC互连的机群系统下，对南京市1个月的浮动车数据进行串行和并行地图匹配对比实验，加速比为1.736。.本项目还对浮动车技术在智能交通领域的数据挖掘展开了较深层次的应用研究。在获取车辆连续轨迹及轨迹上任意点时间的基础上，给出了一种以时空路径为枢纽的车辆时空相遇关系分析方法，结合空间关系理论和Allen的时间段逻辑模型，探讨不同时态关系车辆的相遇关系情况。通过对大规模浮动车轨迹点的数据挖掘，给出了一种利用数学形态学和数字图像处理等方法对道路网进行变化检测和更新的方法。结合三段式速度—流量模型，估算各路段交通流量权值，将其引入非直线系数的计算模型中，使其更符合现实交通需要，结合非直线系数给出路网合理性分析及优化方法，为交通规划提供重要依据。.资助项目执行期间，项目组共发表学术论文6篇，其中EI收录论文5篇。申请计算机软件著作权1项，出版学术专著1部，正在申请国家发明专利2项。.项目组积极推广项目研究成果，先后与四川地信中心、四川九洲导航、江苏智通在内的多家企事业单位围绕技术转化和成果应用开展交流，并在四川地理空间信息大数据中心和四川地理空间数据交换共享平台建设中部分应用项目成果。